Next Gen GPU's will be even more expensive

[chrcoluk]

It's not even the first time I've heard someone say 'nice mid-tier hardware is capable of RT well get RT only games'. That was 6 years ago now and we are still waiting. I'd MUCH rather have ray-traced sound, much less resource intensive and extremely beneficial. Even better, dynamically generate sounds based on how sound propogation work. Every pot should make a different sound based on the weapon hitting it, where it hits it, ect. Literally anything other than graphics for once. Decent AI for example sure would be nice. Modern AAA games are as wide as an ocean and as shallow as a puddle.

According to Nvidia % wise there is a fair jump in capability (at least on xx90) but games as usual are also adding extra intensity to what they releasing and with decreasing levels of optimisation in line with GPUs being more powerful, if I remember correctly the original bastion of RT Cyberpunk 2077, isnt even comparable now, as I think they upgraded the basic RT they originally had to path tracing or something with extra demands on the hardware. Would have to force steam to download the first public build of the game to see on the actual progress there with newer GPUs. Ultimately RT was and probably still is in its orphanage on consumer hardware, and so games are having to only implement it at low levels, and every time a newer gen of hardware comes out they add a bit more to counter that and we will keep going until hardware out paces software, if that even ever happens.

 

[SL2]

Can amd deliver that? Idk.

Yeah, but it's more likely they'll do that than succeed with a premium model. Like I said, the RX 7800 is their best selling current model by far on amz, and that's what they're aiming for. No point in making premium models that doesn't sell well enough.

Also, the ray tracing argument for Nvidia becomes less of a selling point the lower the model number, because even when it's faster than AMD it's still not fast enough. On the other hand, how many of the average buyers understands this lol...

 

[iameatingjam]

Yeah, but it's more likely they'll do that than succeed with a premium model. Like I said, the RX 7800 is their best selling current model by far on amz, and that's what they're aiming for. No point in making premium models that doesn't sell well enough.

Also, the ray tracing argument for Nvidia becomes less of a selling point the lower the model number, because even when it's faster than AMD it's still not fast enough. On the other hand, how many of the average buyers understands this lol...

Yeah no I am completely agreeing with you, I just think it'll be hard either way. At least where I live, the stores don't even stock amd cards, if they do, they might have like one model next 20 different nvidia cards. So it will have to be really attention grabbing, to the point where the info reaches out of the enthusiast space.

 

[AusWolf]

It's mostly annoying as I hate stagnation but that applies to a lot of things for me not just gpu's.

I don't hate stagnation, either, as long as what I have is good enough. "If it's not broken, don't fix it" is a proverb I live by (although it doesn't really show on my hardware purchases, but still).

If everyone gave up making desktop gpus tomorrow I'd happily game on a console but I expect the same, that there is progress each generation....

I actually game on both PS5 and Switch and plan on getting a PS6 and Switch 2.

My hobby is still gaming with the gpu just a small but very expensive portion of that lol. It also isn't my most expensive hobby. Nvidia is trying their hardest to change that though lol....

Consoles aren't my thing. My only consoles are a PS2 and a PSP, although I'm tempted to buy a Steam Deck. If everyone gave up on making desktops, I'd just keep on playing on what I currently have. My backlog is probably long enough to keep me occupied for a lifetime.

 

[oxrufiioxo]

I don't hate stagnation, either, as long as what I have is good enough. "If it's not broken, don't fix it" is a proverb I live by (although it doesn't really show on my hardware purchases, but still).

That's probably the main difference between our views on this even if my gpu couldn't handle games anymore and my financial means dictated I couldn't upgrade I'd still be annoyed if progress stagnated lol.

 

[AusWolf]

That's probably the main difference between our views on this even if my gpu couldn't handle games anymore and my financial means dictated I couldn't upgrade I'd still be annoyed if progress stagnated lol.

Well, if my GPU can't handle games anymore, that's bad. But if it can, then why should I care for a little gap in my upgrade cycle? That's more money saved (that is, spent on other stuff) for a couple more years.

 

[iameatingjam]

I don't hate stagnation, either, as long as what I have is good enough. "If it's not broken, don't fix it" is a proverb I live by.


Consoles aren't my thing. My only consoles are a PS2 and a PSP, although I'm tempted to buy a Steam Deck. If everyone gave up on making desktops, I'd just keep on playing on what I currently have. My backlog is probably long enough to keep me occupied for a lifetime.

I think expecting infinite progress is unrealistic. I mean just look at cars, and how much they improved the first 50 years. But these days, your average ICE car, year to year, aint much different, I mean they got some bells and whistles like GPS and backup cameras and what have you... but the progress of the traditional ice engine has slowed significantly compared to the early years.

At the very least, I think the kind of gains we had in the 90s and 00s, and even the 2010s without price increases, aint gonna happen forever and we're already seeing sings. There'd have to be a major rethink or something (like with electric cars for instance, but even then, there's tradeoffs)

Maybe the focus will be put back onto tight and efficient programming for the best performance. Almost like a complete reversal of how it is now.

Anyway, just thinking out loud.

 

[oxrufiioxo]

I think expecting infinite progress in unrealistic. I mean just look at cars, and how much they improved the first 50 years. But these days, your average ICE car, year to year, aint much different, I mean they got some bells and whistles like GPS and backup cameras and what have you... but the progress of the traditional ice engine has slowed significantly compared to the early years.

At the very least, I think the kind of gains we had in the 90s and 00s, and even the 2010s without price increases, aint gonna happen forever and we're already seeing sings. There'd have to be a major rethink or something (like with electric cars for instance, but even then, there's tradeoffs)

Maybe the focus will be put back onto tight and efficient programming for the best performance. Almost like a complete reversal of how it is now.

Anyway, just thinking out loud.

Yeah, the next 5-10 years will be interesting, and while I think there will be progress it's likely gonna cost a substantial amount more than it does now. They already pretty much abandoned the sub 400 usd range with progress really stagnating there already. Hopefully RDNA4 and RTX 5000 change that at least.

Amd has already stated they need to go for the 80% TAM recently in a interview Hopefully that means actually good products in the 250-400 range.

AMD deprioritizing flagship gaming GPUs: Jack Huynh talks new strategy against Nvidia in gaming market

The battle seems to be over before it starts.

www.tomshardware.com

 

[evernessince]

According to Nvidia % wise there is a fair jump in capability (at least on xx90) but games as usual are also adding extra intensity to what they releasing and with decreasing levels of optimisation in line with GPUs being more powerful, if I remember correctly the original bastion of RT Cyberpunk 2077, isnt even comparable now, as I think they upgraded the basic RT they originally had to path tracing or something with extra demands on the hardware. Would have to force steam to download the first public build of the game to see on the actual progress there with newer GPUs. Ultimately RT was and probably still is in its orphanage on consumer hardware, and so games are having to only implement it at low levels, and every time a newer gen of hardware comes out they add a bit more to counter that and we will keep going until hardware out paces software, if that even ever happens.

I haven't played CP2077 is a hot minute but when they did add path tracing, you were still able to use regular RT. It was up to you which you wanted. Unless that's changed?

I wasn't a fan of the Path-tracing in CP2077, it had a doll house look to it.

Well I actually hope you're right about that one. I don't like tracing and I avoid using it to save power. All I was saying is, I think this is the direction the AAA games are going. Like I said I can't predict the future. Its just my educated guess. Moreso that software raytracing will be simplified and optimized than mid tier hardware will get powerful enough to do what we consider ray tracing today, but it will probably be a bit of both.

Oh for sure, RT is the future. I suspect it'll be much like we have now where devs have access to a variety of tools and techniques each with their own quality level, use case, and performance hit. For example, it might make sense for a dev to use RT GI (global illumination) for an outdoor temple level but due to performance requirements may want to use SSR / planar reflections for the pools of water on the ground. It might make more sense too given pools of water outside a abandoned temple are going to be dirty. You don't want to give people that tech demo effect where just become you can make super detailed reflections doesn't mean you should or that it's realistic.

Yeah, the next 5-10 years will be interesting, and while I think there will be progress it's likely gonna cost a substantial amount more than it does now. They already pretty much abandoned the sub 400 usd range with progress really stagnating there already. Hopefully RDNA4 and RTX 5000 change that at least.

Amd has already stated they need to go for the 80% TAM recently in a interview Hopefully that means actually good products in the 250-400 range.

You have to wonder if modern computers could use a separate accelerator card, perhaps specifically for RT / AI. It's been a very very long time since the paradigm has changed and offloading functions to a separate part would result in much better yields (and thus cost).

 

[oxrufiioxo]

You have to wonder if modern computers could use a separate accelerator card, perhaps specifically for RT / AI. It's been a very very long time since the paradigm has changed and offloading functions to a separate part would result in much better yields (and thus cost).

I think Nvidia/AMD will figure out chiplet gpus with multiple compute dies for gaming before a separate physX like card is a thing.

I could see a dedicated accelerator on a gpu Similar to how consoles have bespoke hardware that handles decompression though.

Either way it's going to be a long time before we have True Pathtracing in a game with no fallbacks.

Hopefully whatever RDNA 4/5 introduces RT wise is a massive jump for games that at least use the current iteration of Pathtracing/full RT even though we are still at the beginning stages.

 

[TumbleGeorge]

Maybe the focus will be put back onto tight and efficient programming for the best performance. Almost like a complete reversal of how it is now.

Vain hopes if they program people. If LLM is used then it is quite possible. As long as capital decides that this is the way. But how would they make us buy further components with increased performance and capacity if AAA games start to be collected on a regular CD again and can be played with a graphics card of the rank of GT 1030?...

 

[tfdsaf]

If AMD want to be successful with their low to mid tier strategy they need to significantly lower prices over the current models. We need to see RX 7900XT type performance from the RX 8800XT at $500 dollars/euros, otherwise that is the point?

The 7800XT type performance needs to cost $300, rather than the $480-500 that it is now, and they need to have a solid $200 offering as well with 12GB of vram and faster than the 7600.

 

[tussinman]

If AMD want to be successful with their low to mid tier strategy they need to significantly lower prices over the current models. We need to see RX 7900XT type performance from the RX 8800XT at $500 dollars/euros, otherwise that is the point?

The 7800XT type performance needs to cost $300, rather than the $480-500 that it is now, and they need to have a solid $200 offering as well with 12GB of vram and faster than the 7600.

That's the issue this gen had. The 7800XT had the same performance as the 6800XT and was $500 which was what the 6800XT was selling for at the time. 7600 was basically a more power efficient 6600XT and once again similar price at the time. 7600XT was slower than the 6700XT.

All the talk was how RDNA 3 was much cheaper to produce than the RDNA 2, but the prices and performance didn't really show that

 

[iameatingjam]

That's the issue this gen had. The 7800XT had the same performance as the 6800XT and was $500 which was what the 6800XT was selling for at the time. 7600 was basically a more power efficient 6600XT and once again similar price at the time. 7600XT was slower than the 6700XT.

All the talk was how RDNA 3 was much cheaper to produce than the RDNA 2, but the prices and performance didn't really show that

I think both nvidia and amd planned to ride the crypto wave for a little longer than it ended up lasting. But we'll see next gen how true that is. Nvidia will be overpriced obviously, just look at their margins. But will the 80 card be $1200? I would guess no.

 

[chrcoluk]

I haven't played CP2077 is a hot minute but when they did add path tracing, you were still able to use regular RT. It was up to you which you wanted. Unless that's changed?

I wasn't a fan of the Path-tracing in CP2077, it had a doll house look to it.

No idea, if you remember it being that way I assume you are correct its optional, but I do recall reading comments from users that when they have downloaded new builds of the game it slows down on their systems, whether thats because defaults are changed to heavier settings or something else who knows, the safest way would probably be still test the first build that was used on reviews of 2000 series cards.

 

[las]

Or they have to repeat what they've done a decade ago with Maxwell. Same node, much more bang per watt compared to Kepler.

However, I don't expect anything but more of the same: stupid high prices and very small performance improvements.

Not going to happen as Ada / 4000 series is doing really well and GTX 700 series sucked compared to 600 and 900 which sold in huge numbers in comparison.

Nvidia can do whatever they want, AMD is sleeping. They are years behind and RDNA4 won't change a thing. They left high-end long ago, first admitting it now tho.

Nvidia's grasp of desktop GPU market balloons to 88% — AMD has just 12%, Intel negligible, says JPR

The discrete GPU industry is returning to seasonality.

www.tomshardware.com

AMD is CPU first and probably uses like 1% of R&D funds on gaming GPUs, obviously they are not going to be a threat to Nvidia which is GPU first.

And by then Nvidia has something faster again....

Yes RTX 6000 series probably hit Q4 2026 or Q1 2027, around 2 years after Blackwell, like usual.

Blackwell looks like a stop gap solution tho. 4N to 4N. I expect RTX 6000 to use TSMC N3 minimum, probably N2.

Might skip 5000 series because of this. Lets see. If 5090 goes to 550-600 watts, I don't think I will be interrested. Too much heat dissipation. Not really going to bother with heavy undervolting when my 4090 runs pretty much anything with ease. RTX 6080 might be the better upgrade eventually. I don't expect 5080 to beat 4090 much so it won't be a solution for me.

 

[TumbleGeorge]

I expect RTX 6000 to use TSMC N3 minimum, probably N2.

The N2 won't be ready in time for such large and complex chips for the RTX 6000 series, not to mention that it will be primarily busy for Apple at first. But the RTX 7000 series is quite likely to use it.

 

[Macro Device]

Not going to happen as Ada / 4000 series is doing really well and GTX 700 series sucked compared to 600 and 900 which sold in huge numbers in comparison.

Why don't you ever try to actually read...

Maxwell (GTX 750 series + GTX 900 series) was built on the same TSMC 28 nm node as any Kepler device (GTX 600 and 700 series excluding GTX 750 series) and had a much better performance per watt ratio. I never made any claims regarding market success.

 

[las]

Why don't you ever try to actually read...

Maxwell (GTX 750 series + GTX 900 series) was built on the same TSMC 28 nm node as any Kepler device (GTX 600 and 700 series excluding GTX 750 series) and had a much better performance per watt ratio. I never made any claims regarding market success.

Why don't you read Blackwell arch deep dive and you will see that the arch has not changed much - You won't see massive gains from the arch alone. GDDR7 will bring some tho, however, probably only if bandwidth is an issue.

Blackwell won't deliver huge gains without a much bigger die or with much higher power usage, 3000 to 4000 was a somewhat big step forward in terms of performance per watt only because TSMC 4N is massively better than Samsung 8nm, which is a renamed 10nm node much closer to TSMC 12nm in reality, but it was cheap and enough to beat AMD which used a superior node in TSMC 7nm - and still lost.

And this is also why 5090 is rumoured to get like up to 25.000 cores which is 50% more than 4090

5090 is probably going to be a beast, maybe MCM, maybe not, however watt usage will be high and price will be too.

 

[AusWolf]

Why don't you read Blackwell arch deep dive and you will see that the arch has not changed much

Blackwell won't deliver huge gains without a much bigger die or with much higher power usage, 3000 to 4000 was a somewhat big step forward in terms of performance per watt only because TSMC 4N is massively better than Samsung 8nm, which is a renamed 10nm node much closer to TSMC 12nm in reality, but it was cheap.

And this is also why 5090 is rumoured to get like up to 25.000 cores which is 50% more than 4090

Nvidia changed almost nothing in the architecture with Ada, and if this continues with Blackwell, then all advancements (or the lack of them) will likely go hand in hand with node shrinks.

 

[las]

Nvidia changed almost nothing in the architecture with Ada, and if this continues with Blackwell, then all advancements (or the lack of them) will likely go hand in hand with node shrinks.

Ada brought huge Power and Efficiency gains over Ampere, which resulted in like 1 GHz higher clockspeeds on avg which is 50% higher. Shows how gimped 3000 series actually were and very powerlimited due to cheap Samsung 8/10nm really.

Also Ada has newer and better tensor cores, better optical flow accelerator which allows for proper Frame Gen (stutters on Ampere, even in modded DLSS FG games, simply don't work with Ampere), better RT, DLSS 3.x, Shader Execution Reordering so yeah, there is architectural improvements in Ada for sure. Many people have tried to run DLSS 3 FG on 3000 series on hacked drivers and they all failed. Nvidia is NOT doing it on purpose, as Ampere simply lacks the capability.

But sure, the biggest improvement was really Samsung 8/10nm to TSMC 4N and Blackwell is not going to get this advantage, because its 4N again, meaning chip size and/or power has to increase to gain alot of performance. GDDR7 brings some performance as well (but might drive prices up even more and Nvidia is rumoured to use only 28Gbps chips even tho GDDR7 can do 32Gbps now)

Nvidia will probably use 32-36 Gbps GDDR7 on 5000 refresh.

Most rumours so far also claim 5080 is 350-400 watts and 5090 is 550-600 watts which fits what I claim; Bigger die / More power, because node shrink won't be a thing for 5000 series. Luckily Nvidia don't really need a huge performance gain for this gen, because AMD left high-end gaming GPU market.

 

[AusWolf]

Ada brought huge Power and Efficiency gains over Ampere, which resulted in like 1 GHz higher clockspeeds on avg which is 50% higher. Shows how gimped 3000 series actually were and very powerlimited due to cheap Samsung 8nm really.

Also Ada has newer and better tensor cores, better optical flow accelerator which allows for proper Frame Gen (stutters on Ampere, even in modded DLSS FG games, simply don't work with Ampere), better RT, DLSS 3.x, Shader Execution Reordering so yeah, there is architectural improvements in Ada for sure.

But sure, the biggest improvement was really Samsung 8nm to TSMC 4N and Blackwell is not going to get this advantage, because its still 4N, meaning chip size and/or power has to increase to gain alot of performance. GDDR7 brings some performance as well (but might drive prices up even more and Nvidia is rumoured to use only 28Gbps chips even tho GDDR7 can do 32Gbps now)

Yeah, Ada got the better optical flow thingy for FG, but I don't see any other advancement, honestly. Enabling RT results in the same performance drop as it did on Ampere or Turing. The efficiency and clock speeds gains are due to the node shrink, not the architecture.

This is the difference between the Nvidia and AMD mindset. If Nvidia says they've got some cool new stuff, everybody praises them for it without checking out how (or if) it works in real life first. If AMD says the same, everybody mocks them for not delivering 20% better everything than Nvidia for a 50% lower price.

 

[las]

The N2 won't be ready in time for such large and complex chips for the RTX 6000 series, not to mention that it will be primarily busy for Apple at first. But the RTX 7000 series is quite likely to use it.

RTX 6000 is 2-2½ years away, you don't really know

Also, Nvidia might be using Intel 18A, 16A, 14A or Samsung by then.

I could easily see Nvidia drop gaming GPUs to a worse and cheaper node eventually, there's little competition especially in the higher end and RDNA4 with Radeon 8000 series won't change this

Nvidias focus is on AI and Enterprise and they saw with 3000 series that they could easily beat AMD in the gaming market using a cheap process

Nvidia will be using TSMC 3N in 2025 for AI and Enterprise already, like 1 year before any of their gaming chips >might< use 3N

Yeah, Ada got the better optical flow thingy for FG, but I don't see any other advancement, honestly. Enabling RT results in the same performance drop as it did on Ampere or Turing. The efficiency and clock speeds gains are due to the node shrink, not the architecture.

This is the difference between the Nvidia and AMD mindset. If Nvidia says they've got some cool new stuff, everybody praises them for it without checking out how (or if) it works in real life first. If AMD says the same, everybody mocks them for not delivering 20% better everything than Nvidia for a 50% lower price.

Many people tried to make DLSS 3 FG work on Ampere but no-one succeded, because architecture can't do it. Try google it. Some people got FG to work, but it was stuttering due to optical flow accelerator being too slow. SER is also an archtectural improvement really. Look up chip designs side by side and you will see they are not that close.

Most make fun of AMD because they are hyping stuff that rarely work. Anti Lag + for example which got people VAC banned on Steam, like they did not even care to test the feature before releasing it. AMD tries to copy/paste RTX features these days, they invent little to nothing new, sadly.

 

[Macro Device]

which resulted in like 1 GHz higher clockspeeds on avg which is 50% higher.

This isn't a measure of power efficiency. What is is you now have a 200 W GPU achieving roughly the same a 320 W of previous generation did.

Most make fun of AMD because they are hyping stuff that rarely work.

For a good reason. This "we also got this but five years later and in a worse shape" approach doesn't really attract. You can't hate them enough.

 

[TumbleGeorge]

RTX 6000 is 2-2½ years away, you don't really know

I'm quite sure for that manufacturing of GPU's for RTX 6000 will beginning almost of year before start sales of graphic cards with it. Too early for Rubin to be on "2"nm.